A novel real-time online defect detection, grading, and damage assessment system for steel pipelines and its intelligent applications

Abstract

In non-destructive testing of steel pipelines, the detection signal is often submerged by the primary field. Achieving weak and reliable identification and effective separation in strongly coupled signals is a key technical challenge for detecting harmonic magnetic fields in buried pipeline damage. Here, we propose a new scheme for real-time online defect detection and grading of steel pipelines. This scheme is based on harmonic excitation and full bridge circuit construction, utilizing the inner product between harmonics at a spatial distance to reconstruct the matching features of the relevant layers for analysis, and inverse performing the corresponding defect level classification method. In addition, a low-latency real-time online processing system was developed by using the simplified form of the analytical solution of this method and the dynamic linking model between the probe movement speed and the processing step size. Finally, an intelligent evaluation based on polar coordinate image transformation and spatial attention mechanism was implemented using the ResNet101 residual network, with higher accuracy (average accuracy of 98.28 %), and a corresponding Edge AI system was established. By conducting pipeline experiments under different working conditions and thinning of different wall thicknesses, a detailed analysis of the quantitative evaluation of pipeline defects was carried out, and the improvement of this scheme for real-time online detection and defect characterization under various complex working conditions was reported. The proposed method has shown potential application value in real-time non-destructive testing, quantitative analysis of pipeline inspection with large lift-off probes, and intelligent early warning.